Appendix 3: Additional models results

1. Models with traits

Specification of the models: We used lme4 package to perform a GLMM with binomial (proportion) distribution. An example of the code for each dataset are as follows:

mhigh.spe <- glmer(cbind(occor, n.visit-occor) ~  
                    forest_site400*lbody_size + 
                    forest_site400*nest +
                    forest_site400*diet + 
                    forest_site400*lower_stratum +
                    forest_land*lbody_size + 
                    forest_land*nest + 
                    forest_land*diet +
                    forest_land*lower_stratum + 
                    (forest_site400 + forest_land|sp) + 
                    (1|landscape:sp) + (1|site:sp) + 
                    (lbody_size + nest + diet + lower_stratum|landscape) +
                    (lbody_size + nest + diet + lower_stratum|site),
                    family=binomial, data=high.spe,
                    nAGQ = 1, control = glmerControl(optimizer = "bobyqa",
                                    optCtrl = list(maxfun = 500000)))

We ran separate models for each assemblage and trait. Afterwards, we ran one model with the combination of the traits body mass, diet, nest type and % of lower strata use. Table S3. shows the marginal R2 of all models terms.

2. Models coeficients

Tables S3., S3., S3., and S3. show the coefficients for each model.

3. Models diagnostic

Variance Inflation Factor of the model parameters for each dataset in Table S3..

Example of the residual diagnostic of the model with the combined traits (main diet, body mass, nest type and % of lower strata use) for the forest specialists in high-quality matrix landscapes. The models’ diagnostics for the other assemblages were all similar and can be checked in this Rmd file.

Residual correlations among species and sites

Below we present the Kendall correlations for the residuals among species and sites for the models using the predictions for site:sp random effect (Observation Level Random Effect). For the residual correlations we followed the code provided by Miller, Damschen & Ives (2018).

Range of species correlations: -0.4, 0.43. Range of sites correlations: -0.3, 0.27.

Species residual Kendall correlations for the specialist species in high-quality matrix landscapes.

Species residual Kendall correlations for the specialist species in high-quality matrix landscapes.

Sites residual Kendall correlations for the specialist species in high-quality matrixlandscapes.

Sites residual Kendall correlations for the specialist species in high-quality matrixlandscapes.

Histograms of the residual Kendall correlations for the specialists species in high-quality matrix landscapes.

Histograms of the residual Kendall correlations for the specialists species in high-quality matrix landscapes.

Residual diagnostic

We used DHARMa package (Hartig (2018)) for the diagnostic of quantile residuals.

Plots for model diagnostic form DHARMa package.

Plots for model diagnostic form DHARMa package.

Residuals against predictors:

More for model diagnostic form DHARMa package.

More for model diagnostic form DHARMa package.

Predictions for each species local forest cover

Landscape forest cover was fixed in 30%.

Forest specialist birds.

Forest specialist birds.

Forest specialist birds.

Forest specialist birds.

References

Hartig, F. (2018). DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models.
Miller, J.E.D., Damschen, E.I. & Ives, A.R. (2018). Functional traits and community composition: A comparison among community-weighted means, weighted correlations, and multilevel models. Methods in Ecology and Evolution 0.